Real-Time Object Detection And Localization In Compressive Sensed Video

Typically a 1-2MP CCTV camera generates around 7-12GBof data per day. Frame-by-frame processing of such an enormous amount of data requires hefty computational resources. In recent years, compressive sensing approaches have shown impressive signal processing results by reducing the sampling bandwidth. Different sampling mechanisms were developed to incorporate compressive sensing in image and video acquisition. Though all-CMOS sensor cameras that perform compressive sensing save a lot of bandwidth on sampling and minimize the memory required to store videos. However, traditional signal processing and deep learning model can realize operations only on the reconstructed data. The reconstruction of compressive-sensed videos is computationally expensive and time-consuming. In this work, we propose a sparse deep learning model to overcome this overhead to detect and localize the objects directly on the compressed frames. Thus, mitigating the need to reconstruct the frames and reducing the search rate up to 20 times (compression rate). We achieved an accuracy of 46.27% mAP with the proposed model on GeForce GTX 1080 Ti. We were also able to show real-time inference on an NVIDIA TX2 embedded board with 45.11%mAP.